gms | German Medical Science

Objective: Atypical meningiomas (WHO grade II) comprise a relatively heterogeneous tumor entity. New molecular markers, associated with tumor aggressiveness have been recently introduced, whereas radiological criteria remain less defined.

Recently, we found that integration of radiomic and semantic features might be a promising tool to identify high-risk atypical meningiomas. In this study, we sought to evaluate the potential value of available histopathological data for the performance of this predictor model.

Methods: Primary atypical meningiomas with preoperative MRIs operated from 2003 to 2017 in our department were included in the study (n=76). Homogeneous radiomic data for contrast-enhanced T1 sequences was available for 52 patients. 25 radiomic, 11 histopathologic and 11 semantic features, potentially associated with tumor aggressiveness and clinical data including progression-free survival were used. Multiple imputation for the missing values was performed. We used univariate and multivariate regression for the outcome analysis and AUC for feature prediction.

Results: Mean age was 58,7 (SD 13,8) years, there were 59,2% women, the majority of tumors were localized on the convexity and falx, 13,2% could be resected completely (Simpson grade 1 or 2). Tumor relapse occurred in 22,4% of cases.

We found a predictive potential for Ki67 index (AUC 0,723, p=0,02 (difference entropy)) and brain invasion (AUC 0,706, p=0,004 (minimum)) using certain semantic characteristics. No association was detected between semantic and histological features. There were no tumor relapses occurring in cases with 5 or fewer mitoses per field. However, short follow up time must be taken into account. High cluster prominence was associated with tumor relapse (HR 5,9 (1,03-33,73)). High cellularity (HR 3,6 (95% CI 1,01-12,84)) and Cystic component (HR 9,77, 95%CI 3,14-30,41) were associated with shorter PFS. However, adding these histological criteria to the predictory model did not increase the classification power (AUC 0,853 vs. 0,855).

Conclusion: In this study, certain histological characteristics were associated with preoperative radiomic features. These additional information might be valuable for surgical strategy. Due to high classification power of integrated radiomic and semantic model, the histological criteria added little value.